File PACKAGING-NOTES of Package python3
Notes for packagers of Python3
0. Faster build turnaround
By default, python builds with profile-guided optimization. This needs
an additional run of the test suite and it is generally slow.
PGO build takes around 50 minutes.
For development, use "--without profileopt" option to disable PGO. This
shortens the build time to ~5 minutes including test suite.
If you just need to rerun a build and don't need the full test suite, you can
also build "--without testsuite". This shaves off another three minutes.
(Note that you usually *want* the full test suite.)
I. python3 and python3-base naming confusion
1. the important stuff first
This is package "python3". However, master spec file for this package is
"python3-base.spec". That means that all important changes should be put into
"python3-base.spec" and then submitted against "python3" in OBS.
Changelogs of python3-base and python3 will be merged at some point. Now they
aren't, but feel free to enter changes in either or both.
2. why is that?
Technical reasons. python3-base was originally supposed to be a minimal package
with no external dependencies - so that it can build early in the distribution
rebuild. There were also some build loops involved. Turns out, 90% of Python's
standard library can be built without external dependencies. That's what we do -
in python3-base. python3 then only contains the remaining bits -
dependency-heavy subpackages, plus small bits depending on common packages
(OpenSSL, expat, readline) Logically, python3-base must be the master spec
because it builds the important parts.
3. why not rename the whole package to python3-base?
Because that would be stupid. The package is called python3.
4. so why not make python3 the minimal package and put the rest in
Because other distributions use python3-base as the minimal package as well.
Also, packages that require python3 expect the whole deal, not a stripped-down
5. alright, let's build python3-base from python3.spec and python3 from
Tried that, abandoned it. It is more trouble than it's worth.
Our pre_checkin.sh takes care of copying relevant portions of python3-base.spec
to python3.spec. The "relevant portions" are:
* list of patches list of macro definitions patch apply sequence in %prep
* configure call and build setup for %build
That means that when you're adding a patch, you need only add it to
python3-base.spec and it will appear magically in python3.spec too. (as long as
you remember to run pre_checkin.sh or have automation in place to do it for you)
If you want to replicate another section from py3-base to py3, just mark the
section with NEW-SECTION-BEGIN and NEW-SECTION-END (in both py3-base and py3)
and add NEW-SECTION to list of $sections in pre_checkin.sh
(note however that the order of sections in pre_checkin.sh must be the same as
they are found in the spec files)
pre_checkin.sh also finds the newest python tarball in the source directory and
updates versions in all specfiles to match it. Therefore, you usually don't need
to update version numbers in specs when you update Python
We also regenerate import_failed map and check test exclusion lists.
This is a mechanism installed as part of python3-base, that places shim modules
on python's path (through a generated zzzz-import-failed-hooks.pth file, so that
it is imported as much at the end as makes sense; and an _import_failed subdir
of /usr/lib/pythonX.Y). Then when the user tries to import a module that is part
of a subpackage, the ImportError will contain a helpful message telling them
which missing subpackage to install.
This can sometimes cause problems on non-standard configurations, if the pth
gets included too early (for instance if you are using a script to include all
pths by hand in some strange order). Just something to look out for.
All this is based on information in import_failed.map, which is generated from
spec file filelists by pre_checkin.sh